You're being a sophist by comparing those. It's non-comparable type of periodicity. AI doesn't need continuous training to function. In theory, eveyone could just stop and settle on models we have already and they would continue to work forever. Training models is only needed for creating better improved models. And you can count really big models like ChatGPT ones on fingers. You can't raise beef once and keep people fed forever. Raising beef has the same type of periodicity as running already-trained model, something that is continuously done for industry to function.